Explore passing the input message to the retriever in conversational pipeline

[rchan26]

In the conversational pipeline (see #95), we allow the conversational agent llm to use the retriever like a tool and it decides the query to input to the retriever. In the case where the input is quite informative, we might be losing some information from the model. This issue is to explore how model tends to use the retriever and if it's also useful to pass in the actual message to the retriever too like in the standard RAG pipeline.